1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_PROCESSOR_H
3 #define _ASM_X86_PROCESSOR_H
4
5 #include <asm/processor-flags.h>
6
7 /* Forward declaration, a strange C thing */
8 struct task_struct;
9 struct mm_struct;
10 struct io_bitmap;
11 struct vm86;
12
13 #include <asm/math_emu.h>
14 #include <asm/segment.h>
15 #include <asm/types.h>
16 #include <uapi/asm/sigcontext.h>
17 #include <asm/current.h>
18 #include <asm/cpufeatures.h>
19 #include <asm/page.h>
20 #include <asm/pgtable_types.h>
21 #include <asm/percpu.h>
22 #include <asm/msr.h>
23 #include <asm/desc_defs.h>
24 #include <asm/nops.h>
25 #include <asm/special_insns.h>
26 #include <asm/fpu/types.h>
27 #include <asm/unwind_hints.h>
28 #include <asm/vmxfeatures.h>
29 #include <asm/vdso/processor.h>
30
31 #include <linux/personality.h>
32 #include <linux/cache.h>
33 #include <linux/threads.h>
34 #include <linux/math64.h>
35 #include <linux/err.h>
36 #include <linux/irqflags.h>
37 #include <linux/mem_encrypt.h>
38
39 /*
40 * We handle most unaligned accesses in hardware. On the other hand
41 * unaligned DMA can be quite expensive on some Nehalem processors.
42 *
43 * Based on this we disable the IP header alignment in network drivers.
44 */
45 #define NET_IP_ALIGN 0
46
47 #define HBP_NUM 4
48
49 /*
50 * These alignment constraints are for performance in the vSMP case,
51 * but in the task_struct case we must also meet hardware imposed
52 * alignment requirements of the FPU state:
53 */
54 #ifdef CONFIG_X86_VSMP
55 # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT)
56 # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT)
57 #else
58 # define ARCH_MIN_TASKALIGN __alignof__(union fpregs_state)
59 # define ARCH_MIN_MMSTRUCT_ALIGN 0
60 #endif
61
62 enum tlb_infos {
63 ENTRIES,
64 NR_INFO
65 };
66
67 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
68 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
69 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
70 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
71 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
72 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
73 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
74
75 /*
76 * CPU type and hardware bug flags. Kept separately for each CPU.
77 * Members of this structure are referenced in head_32.S, so think twice
78 * before touching them. [mj]
79 */
80
81 struct cpuinfo_x86 {
82 __u8 x86; /* CPU family */
83 __u8 x86_vendor; /* CPU vendor */
84 __u8 x86_model;
85 __u8 x86_stepping;
86 #ifdef CONFIG_X86_64
87 /* Number of 4K pages in DTLB/ITLB combined(in pages): */
88 int x86_tlbsize;
89 #endif
90 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
91 __u32 vmx_capability[NVMXINTS];
92 #endif
93 __u8 x86_virt_bits;
94 __u8 x86_phys_bits;
95 /* CPUID returned core id bits: */
96 __u8 x86_coreid_bits;
97 __u8 cu_id;
98 /* Max extended CPUID function supported: */
99 __u32 extended_cpuid_level;
100 /* Maximum supported CPUID level, -1=no CPUID: */
101 int cpuid_level;
102 /*
103 * Align to size of unsigned long because the x86_capability array
104 * is passed to bitops which require the alignment. Use unnamed
105 * union to enforce the array is aligned to size of unsigned long.
106 */
107 union {
108 __u32 x86_capability[NCAPINTS + NBUGINTS];
109 unsigned long x86_capability_alignment;
110 };
111 char x86_vendor_id[16];
112 char x86_model_id[64];
113 /* in KB - valid for CPUS which support this call: */
114 unsigned int x86_cache_size;
115 int x86_cache_alignment; /* In bytes */
116 /* Cache QoS architectural values, valid only on the BSP: */
117 int x86_cache_max_rmid; /* max index */
118 int x86_cache_occ_scale; /* scale to bytes */
119 int x86_cache_mbm_width_offset;
120 int x86_power;
121 unsigned long loops_per_jiffy;
122 /* cpuid returned max cores value: */
123 u16 x86_max_cores;
124 u16 apicid;
125 u16 initial_apicid;
126 u16 x86_clflush_size;
127 /* number of cores as seen by the OS: */
128 u16 booted_cores;
129 /* Physical processor id: */
130 u16 phys_proc_id;
131 /* Logical processor id: */
132 u16 logical_proc_id;
133 /* Core id: */
134 u16 cpu_core_id;
135 u16 cpu_die_id;
136 u16 logical_die_id;
137 /* Index into per_cpu list: */
138 u16 cpu_index;
139 /* Is SMT active on this core? */
140 bool smt_active;
141 u32 microcode;
142 /* Address space bits used by the cache internally */
143 u8 x86_cache_bits;
144 unsigned initialized : 1;
145 } __randomize_layout;
146
147 struct cpuid_regs {
148 u32 eax, ebx, ecx, edx;
149 };
150
151 enum cpuid_regs_idx {
152 CPUID_EAX = 0,
153 CPUID_EBX,
154 CPUID_ECX,
155 CPUID_EDX,
156 };
157
158 #define X86_VENDOR_INTEL 0
159 #define X86_VENDOR_CYRIX 1
160 #define X86_VENDOR_AMD 2
161 #define X86_VENDOR_UMC 3
162 #define X86_VENDOR_CENTAUR 5
163 #define X86_VENDOR_TRANSMETA 7
164 #define X86_VENDOR_NSC 8
165 #define X86_VENDOR_HYGON 9
166 #define X86_VENDOR_ZHAOXIN 10
167 #define X86_VENDOR_NUM 11
168
169 #define X86_VENDOR_UNKNOWN 0xff
170
171 /*
172 * capabilities of CPUs
173 */
174 extern struct cpuinfo_x86 boot_cpu_data;
175 extern struct cpuinfo_x86 new_cpu_data;
176
177 extern __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS];
178 extern __u32 cpu_caps_set[NCAPINTS + NBUGINTS];
179
180 #ifdef CONFIG_SMP
181 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
182 #define cpu_data(cpu) per_cpu(cpu_info, cpu)
183 #else
184 #define cpu_info boot_cpu_data
185 #define cpu_data(cpu) boot_cpu_data
186 #endif
187
188 extern const struct seq_operations cpuinfo_op;
189
190 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
191
192 extern void cpu_detect(struct cpuinfo_x86 *c);
193
l1tf_pfn_limit(void)194 static inline unsigned long long l1tf_pfn_limit(void)
195 {
196 return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
197 }
198
199 extern void early_cpu_init(void);
200 extern void identify_boot_cpu(void);
201 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
202 extern void print_cpu_info(struct cpuinfo_x86 *);
203 void print_cpu_msr(struct cpuinfo_x86 *);
204
205 #ifdef CONFIG_X86_32
206 extern int have_cpuid_p(void);
207 #else
have_cpuid_p(void)208 static inline int have_cpuid_p(void)
209 {
210 return 1;
211 }
212 #endif
native_cpuid(unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)213 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
214 unsigned int *ecx, unsigned int *edx)
215 {
216 /* ecx is often an input as well as an output. */
217 asm volatile("cpuid"
218 : "=a" (*eax),
219 "=b" (*ebx),
220 "=c" (*ecx),
221 "=d" (*edx)
222 : "0" (*eax), "2" (*ecx)
223 : "memory");
224 }
225
226 #define native_cpuid_reg(reg) \
227 static inline unsigned int native_cpuid_##reg(unsigned int op) \
228 { \
229 unsigned int eax = op, ebx, ecx = 0, edx; \
230 \
231 native_cpuid(&eax, &ebx, &ecx, &edx); \
232 \
233 return reg; \
234 }
235
236 /*
237 * Native CPUID functions returning a single datum.
238 */
239 native_cpuid_reg(eax)
native_cpuid_reg(ebx)240 native_cpuid_reg(ebx)
241 native_cpuid_reg(ecx)
242 native_cpuid_reg(edx)
243
244 /*
245 * Friendlier CR3 helpers.
246 */
247 static inline unsigned long read_cr3_pa(void)
248 {
249 return __read_cr3() & CR3_ADDR_MASK;
250 }
251
native_read_cr3_pa(void)252 static inline unsigned long native_read_cr3_pa(void)
253 {
254 return __native_read_cr3() & CR3_ADDR_MASK;
255 }
256
load_cr3(pgd_t * pgdir)257 static inline void load_cr3(pgd_t *pgdir)
258 {
259 write_cr3(__sme_pa(pgdir));
260 }
261
262 /*
263 * Note that while the legacy 'TSS' name comes from 'Task State Segment',
264 * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
265 * unrelated to the task-switch mechanism:
266 */
267 #ifdef CONFIG_X86_32
268 /* This is the TSS defined by the hardware. */
269 struct x86_hw_tss {
270 unsigned short back_link, __blh;
271 unsigned long sp0;
272 unsigned short ss0, __ss0h;
273 unsigned long sp1;
274
275 /*
276 * We don't use ring 1, so ss1 is a convenient scratch space in
277 * the same cacheline as sp0. We use ss1 to cache the value in
278 * MSR_IA32_SYSENTER_CS. When we context switch
279 * MSR_IA32_SYSENTER_CS, we first check if the new value being
280 * written matches ss1, and, if it's not, then we wrmsr the new
281 * value and update ss1.
282 *
283 * The only reason we context switch MSR_IA32_SYSENTER_CS is
284 * that we set it to zero in vm86 tasks to avoid corrupting the
285 * stack if we were to go through the sysenter path from vm86
286 * mode.
287 */
288 unsigned short ss1; /* MSR_IA32_SYSENTER_CS */
289
290 unsigned short __ss1h;
291 unsigned long sp2;
292 unsigned short ss2, __ss2h;
293 unsigned long __cr3;
294 unsigned long ip;
295 unsigned long flags;
296 unsigned long ax;
297 unsigned long cx;
298 unsigned long dx;
299 unsigned long bx;
300 unsigned long sp;
301 unsigned long bp;
302 unsigned long si;
303 unsigned long di;
304 unsigned short es, __esh;
305 unsigned short cs, __csh;
306 unsigned short ss, __ssh;
307 unsigned short ds, __dsh;
308 unsigned short fs, __fsh;
309 unsigned short gs, __gsh;
310 unsigned short ldt, __ldth;
311 unsigned short trace;
312 unsigned short io_bitmap_base;
313
314 } __attribute__((packed));
315 #else
316 struct x86_hw_tss {
317 u32 reserved1;
318 u64 sp0;
319 u64 sp1;
320
321 /*
322 * Since Linux does not use ring 2, the 'sp2' slot is unused by
323 * hardware. entry_SYSCALL_64 uses it as scratch space to stash
324 * the user RSP value.
325 */
326 u64 sp2;
327
328 u64 reserved2;
329 u64 ist[7];
330 u32 reserved3;
331 u32 reserved4;
332 u16 reserved5;
333 u16 io_bitmap_base;
334
335 } __attribute__((packed));
336 #endif
337
338 /*
339 * IO-bitmap sizes:
340 */
341 #define IO_BITMAP_BITS 65536
342 #define IO_BITMAP_BYTES (IO_BITMAP_BITS / BITS_PER_BYTE)
343 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES / sizeof(long))
344
345 #define IO_BITMAP_OFFSET_VALID_MAP \
346 (offsetof(struct tss_struct, io_bitmap.bitmap) - \
347 offsetof(struct tss_struct, x86_tss))
348
349 #define IO_BITMAP_OFFSET_VALID_ALL \
350 (offsetof(struct tss_struct, io_bitmap.mapall) - \
351 offsetof(struct tss_struct, x86_tss))
352
353 #ifdef CONFIG_X86_IOPL_IOPERM
354 /*
355 * sizeof(unsigned long) coming from an extra "long" at the end of the
356 * iobitmap. The limit is inclusive, i.e. the last valid byte.
357 */
358 # define __KERNEL_TSS_LIMIT \
359 (IO_BITMAP_OFFSET_VALID_ALL + IO_BITMAP_BYTES + \
360 sizeof(unsigned long) - 1)
361 #else
362 # define __KERNEL_TSS_LIMIT \
363 (offsetof(struct tss_struct, x86_tss) + sizeof(struct x86_hw_tss) - 1)
364 #endif
365
366 /* Base offset outside of TSS_LIMIT so unpriviledged IO causes #GP */
367 #define IO_BITMAP_OFFSET_INVALID (__KERNEL_TSS_LIMIT + 1)
368
369 struct entry_stack {
370 char stack[PAGE_SIZE];
371 };
372
373 struct entry_stack_page {
374 struct entry_stack stack;
375 } __aligned(PAGE_SIZE);
376
377 /*
378 * All IO bitmap related data stored in the TSS:
379 */
380 struct x86_io_bitmap {
381 /* The sequence number of the last active bitmap. */
382 u64 prev_sequence;
383
384 /*
385 * Store the dirty size of the last io bitmap offender. The next
386 * one will have to do the cleanup as the switch out to a non io
387 * bitmap user will just set x86_tss.io_bitmap_base to a value
388 * outside of the TSS limit. So for sane tasks there is no need to
389 * actually touch the io_bitmap at all.
390 */
391 unsigned int prev_max;
392
393 /*
394 * The extra 1 is there because the CPU will access an
395 * additional byte beyond the end of the IO permission
396 * bitmap. The extra byte must be all 1 bits, and must
397 * be within the limit.
398 */
399 unsigned long bitmap[IO_BITMAP_LONGS + 1];
400
401 /*
402 * Special I/O bitmap to emulate IOPL(3). All bytes zero,
403 * except the additional byte at the end.
404 */
405 unsigned long mapall[IO_BITMAP_LONGS + 1];
406 };
407
408 struct tss_struct {
409 /*
410 * The fixed hardware portion. This must not cross a page boundary
411 * at risk of violating the SDM's advice and potentially triggering
412 * errata.
413 */
414 struct x86_hw_tss x86_tss;
415
416 struct x86_io_bitmap io_bitmap;
417 } __aligned(PAGE_SIZE);
418
419 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
420
421 /* Per CPU interrupt stacks */
422 struct irq_stack {
423 char stack[IRQ_STACK_SIZE];
424 } __aligned(IRQ_STACK_SIZE);
425
426 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
427
428 #ifdef CONFIG_X86_64
429 struct fixed_percpu_data {
430 /*
431 * GCC hardcodes the stack canary as %gs:40. Since the
432 * irq_stack is the object at %gs:0, we reserve the bottom
433 * 48 bytes of the irq stack for the canary.
434 *
435 * Once we are willing to require -mstack-protector-guard-symbol=
436 * support for x86_64 stackprotector, we can get rid of this.
437 */
438 char gs_base[40];
439 unsigned long stack_canary;
440 };
441
442 DECLARE_PER_CPU_FIRST(struct fixed_percpu_data, fixed_percpu_data) __visible;
443 DECLARE_INIT_PER_CPU(fixed_percpu_data);
444
cpu_kernelmode_gs_base(int cpu)445 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
446 {
447 return (unsigned long)per_cpu(fixed_percpu_data.gs_base, cpu);
448 }
449
450 DECLARE_PER_CPU(void *, hardirq_stack_ptr);
451 DECLARE_PER_CPU(bool, hardirq_stack_inuse);
452 extern asmlinkage void ignore_sysret(void);
453
454 /* Save actual FS/GS selectors and bases to current->thread */
455 void current_save_fsgs(void);
456 #else /* X86_64 */
457 #ifdef CONFIG_STACKPROTECTOR
458 DECLARE_PER_CPU(unsigned long, __stack_chk_guard);
459 #endif
460 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
461 DECLARE_PER_CPU(struct irq_stack *, softirq_stack_ptr);
462 #endif /* !X86_64 */
463
464 extern unsigned int fpu_kernel_xstate_size;
465 extern unsigned int fpu_user_xstate_size;
466
467 struct perf_event;
468
469 struct thread_struct {
470 /* Cached TLS descriptors: */
471 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
472 #ifdef CONFIG_X86_32
473 unsigned long sp0;
474 #endif
475 unsigned long sp;
476 #ifdef CONFIG_X86_32
477 unsigned long sysenter_cs;
478 #else
479 unsigned short es;
480 unsigned short ds;
481 unsigned short fsindex;
482 unsigned short gsindex;
483 #endif
484
485 #ifdef CONFIG_X86_64
486 unsigned long fsbase;
487 unsigned long gsbase;
488 #else
489 /*
490 * XXX: this could presumably be unsigned short. Alternatively,
491 * 32-bit kernels could be taught to use fsindex instead.
492 */
493 unsigned long fs;
494 unsigned long gs;
495 #endif
496
497 /* Save middle states of ptrace breakpoints */
498 struct perf_event *ptrace_bps[HBP_NUM];
499 /* Debug status used for traps, single steps, etc... */
500 unsigned long virtual_dr6;
501 /* Keep track of the exact dr7 value set by the user */
502 unsigned long ptrace_dr7;
503 /* Fault info: */
504 unsigned long cr2;
505 unsigned long trap_nr;
506 unsigned long error_code;
507 #ifdef CONFIG_VM86
508 /* Virtual 86 mode info */
509 struct vm86 *vm86;
510 #endif
511 /* IO permissions: */
512 struct io_bitmap *io_bitmap;
513
514 /*
515 * IOPL. Privilege level dependent I/O permission which is
516 * emulated via the I/O bitmap to prevent user space from disabling
517 * interrupts.
518 */
519 unsigned long iopl_emul;
520
521 unsigned int iopl_warn:1;
522 unsigned int sig_on_uaccess_err:1;
523
524 /*
525 * Protection Keys Register for Userspace. Loaded immediately on
526 * context switch. Store it in thread_struct to avoid a lookup in
527 * the tasks's FPU xstate buffer. This value is only valid when a
528 * task is scheduled out. For 'current' the authoritative source of
529 * PKRU is the hardware itself.
530 */
531 u32 pkru;
532
533 /* Floating point and extended processor state */
534 struct fpu fpu;
535 /*
536 * WARNING: 'fpu' is dynamically-sized. It *MUST* be at
537 * the end.
538 */
539 };
540
541 /* Whitelist the FPU state from the task_struct for hardened usercopy. */
arch_thread_struct_whitelist(unsigned long * offset,unsigned long * size)542 static inline void arch_thread_struct_whitelist(unsigned long *offset,
543 unsigned long *size)
544 {
545 *offset = offsetof(struct thread_struct, fpu.state);
546 *size = fpu_kernel_xstate_size;
547 }
548
549 static inline void
native_load_sp0(unsigned long sp0)550 native_load_sp0(unsigned long sp0)
551 {
552 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
553 }
554
native_swapgs(void)555 static __always_inline void native_swapgs(void)
556 {
557 #ifdef CONFIG_X86_64
558 asm volatile("swapgs" ::: "memory");
559 #endif
560 }
561
current_top_of_stack(void)562 static inline unsigned long current_top_of_stack(void)
563 {
564 /*
565 * We can't read directly from tss.sp0: sp0 on x86_32 is special in
566 * and around vm86 mode and sp0 on x86_64 is special because of the
567 * entry trampoline.
568 */
569 return this_cpu_read_stable(cpu_current_top_of_stack);
570 }
571
on_thread_stack(void)572 static inline bool on_thread_stack(void)
573 {
574 return (unsigned long)(current_top_of_stack() -
575 current_stack_pointer) < THREAD_SIZE;
576 }
577
578 #ifdef CONFIG_PARAVIRT_XXL
579 #include <asm/paravirt.h>
580 #else
581 #define __cpuid native_cpuid
582
load_sp0(unsigned long sp0)583 static inline void load_sp0(unsigned long sp0)
584 {
585 native_load_sp0(sp0);
586 }
587
588 #endif /* CONFIG_PARAVIRT_XXL */
589
590 /* Free all resources held by a thread. */
591 extern void release_thread(struct task_struct *);
592
593 unsigned long get_wchan(struct task_struct *p);
594
595 /*
596 * Generic CPUID function
597 * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
598 * resulting in stale register contents being returned.
599 */
cpuid(unsigned int op,unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)600 static inline void cpuid(unsigned int op,
601 unsigned int *eax, unsigned int *ebx,
602 unsigned int *ecx, unsigned int *edx)
603 {
604 *eax = op;
605 *ecx = 0;
606 __cpuid(eax, ebx, ecx, edx);
607 }
608
609 /* Some CPUID calls want 'count' to be placed in ecx */
cpuid_count(unsigned int op,int count,unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)610 static inline void cpuid_count(unsigned int op, int count,
611 unsigned int *eax, unsigned int *ebx,
612 unsigned int *ecx, unsigned int *edx)
613 {
614 *eax = op;
615 *ecx = count;
616 __cpuid(eax, ebx, ecx, edx);
617 }
618
619 /*
620 * CPUID functions returning a single datum
621 */
cpuid_eax(unsigned int op)622 static inline unsigned int cpuid_eax(unsigned int op)
623 {
624 unsigned int eax, ebx, ecx, edx;
625
626 cpuid(op, &eax, &ebx, &ecx, &edx);
627
628 return eax;
629 }
630
cpuid_ebx(unsigned int op)631 static inline unsigned int cpuid_ebx(unsigned int op)
632 {
633 unsigned int eax, ebx, ecx, edx;
634
635 cpuid(op, &eax, &ebx, &ecx, &edx);
636
637 return ebx;
638 }
639
cpuid_ecx(unsigned int op)640 static inline unsigned int cpuid_ecx(unsigned int op)
641 {
642 unsigned int eax, ebx, ecx, edx;
643
644 cpuid(op, &eax, &ebx, &ecx, &edx);
645
646 return ecx;
647 }
648
cpuid_edx(unsigned int op)649 static inline unsigned int cpuid_edx(unsigned int op)
650 {
651 unsigned int eax, ebx, ecx, edx;
652
653 cpuid(op, &eax, &ebx, &ecx, &edx);
654
655 return edx;
656 }
657
658 extern void select_idle_routine(const struct cpuinfo_x86 *c);
659 extern void amd_e400_c1e_apic_setup(void);
660
661 extern unsigned long boot_option_idle_override;
662
663 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
664 IDLE_POLL};
665
666 extern void enable_sep_cpu(void);
667 extern int sysenter_setup(void);
668
669
670 /* Defined in head.S */
671 extern struct desc_ptr early_gdt_descr;
672
673 extern void switch_to_new_gdt(int);
674 extern void load_direct_gdt(int);
675 extern void load_fixmap_gdt(int);
676 extern void load_percpu_segment(int);
677 extern void cpu_init(void);
678 extern void cpu_init_secondary(void);
679 extern void cpu_init_exception_handling(void);
680 extern void cr4_init(void);
681
get_debugctlmsr(void)682 static inline unsigned long get_debugctlmsr(void)
683 {
684 unsigned long debugctlmsr = 0;
685
686 #ifndef CONFIG_X86_DEBUGCTLMSR
687 if (boot_cpu_data.x86 < 6)
688 return 0;
689 #endif
690 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
691
692 return debugctlmsr;
693 }
694
update_debugctlmsr(unsigned long debugctlmsr)695 static inline void update_debugctlmsr(unsigned long debugctlmsr)
696 {
697 #ifndef CONFIG_X86_DEBUGCTLMSR
698 if (boot_cpu_data.x86 < 6)
699 return;
700 #endif
701 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
702 }
703
704 extern void set_task_blockstep(struct task_struct *task, bool on);
705
706 /* Boot loader type from the setup header: */
707 extern int bootloader_type;
708 extern int bootloader_version;
709
710 extern char ignore_fpu_irq;
711
712 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
713 #define ARCH_HAS_PREFETCHW
714 #define ARCH_HAS_SPINLOCK_PREFETCH
715
716 #ifdef CONFIG_X86_32
717 # define BASE_PREFETCH ""
718 # define ARCH_HAS_PREFETCH
719 #else
720 # define BASE_PREFETCH "prefetcht0 %P1"
721 #endif
722
723 /*
724 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
725 *
726 * It's not worth to care about 3dnow prefetches for the K6
727 * because they are microcoded there and very slow.
728 */
prefetch(const void * x)729 static inline void prefetch(const void *x)
730 {
731 alternative_input(BASE_PREFETCH, "prefetchnta %P1",
732 X86_FEATURE_XMM,
733 "m" (*(const char *)x));
734 }
735
736 /*
737 * 3dnow prefetch to get an exclusive cache line.
738 * Useful for spinlocks to avoid one state transition in the
739 * cache coherency protocol:
740 */
prefetchw(const void * x)741 static __always_inline void prefetchw(const void *x)
742 {
743 alternative_input(BASE_PREFETCH, "prefetchw %P1",
744 X86_FEATURE_3DNOWPREFETCH,
745 "m" (*(const char *)x));
746 }
747
spin_lock_prefetch(const void * x)748 static inline void spin_lock_prefetch(const void *x)
749 {
750 prefetchw(x);
751 }
752
753 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
754 TOP_OF_KERNEL_STACK_PADDING)
755
756 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
757
758 #define task_pt_regs(task) \
759 ({ \
760 unsigned long __ptr = (unsigned long)task_stack_page(task); \
761 __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \
762 ((struct pt_regs *)__ptr) - 1; \
763 })
764
765 #ifdef CONFIG_X86_32
766 #define INIT_THREAD { \
767 .sp0 = TOP_OF_INIT_STACK, \
768 .sysenter_cs = __KERNEL_CS, \
769 }
770
771 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
772
773 #else
774 #define INIT_THREAD { }
775
776 extern unsigned long KSTK_ESP(struct task_struct *task);
777
778 #endif /* CONFIG_X86_64 */
779
780 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
781 unsigned long new_sp);
782
783 /*
784 * This decides where the kernel will search for a free chunk of vm
785 * space during mmap's.
786 */
787 #define __TASK_UNMAPPED_BASE(task_size) (PAGE_ALIGN(task_size / 3))
788 #define TASK_UNMAPPED_BASE __TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
789
790 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
791
792 /* Get/set a process' ability to use the timestamp counter instruction */
793 #define GET_TSC_CTL(adr) get_tsc_mode((adr))
794 #define SET_TSC_CTL(val) set_tsc_mode((val))
795
796 extern int get_tsc_mode(unsigned long adr);
797 extern int set_tsc_mode(unsigned int val);
798
799 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
800
801 extern u16 get_llc_id(unsigned int cpu);
802
803 #ifdef CONFIG_CPU_SUP_AMD
804 extern u32 amd_get_nodes_per_socket(void);
805 extern u32 amd_get_highest_perf(void);
806 extern bool cpu_has_ibpb_brtype_microcode(void);
807 extern void amd_clear_divider(void);
808 #else
amd_get_nodes_per_socket(void)809 static inline u32 amd_get_nodes_per_socket(void) { return 0; }
amd_get_highest_perf(void)810 static inline u32 amd_get_highest_perf(void) { return 0; }
cpu_has_ibpb_brtype_microcode(void)811 static inline bool cpu_has_ibpb_brtype_microcode(void) { return false; }
amd_clear_divider(void)812 static inline void amd_clear_divider(void) { }
813 #endif
814
hypervisor_cpuid_base(const char * sig,uint32_t leaves)815 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
816 {
817 uint32_t base, eax, signature[3];
818
819 for (base = 0x40000000; base < 0x40010000; base += 0x100) {
820 cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
821
822 if (!memcmp(sig, signature, 12) &&
823 (leaves == 0 || ((eax - base) >= leaves)))
824 return base;
825 }
826
827 return 0;
828 }
829
830 extern unsigned long arch_align_stack(unsigned long sp);
831 void free_init_pages(const char *what, unsigned long begin, unsigned long end);
832 extern void free_kernel_image_pages(const char *what, void *begin, void *end);
833
834 void default_idle(void);
835 #ifdef CONFIG_XEN
836 bool xen_set_default_idle(void);
837 #else
838 #define xen_set_default_idle 0
839 #endif
840
841 void __noreturn stop_this_cpu(void *dummy);
842 void microcode_check(struct cpuinfo_x86 *prev_info);
843 void store_cpu_caps(struct cpuinfo_x86 *info);
844
845 enum l1tf_mitigations {
846 L1TF_MITIGATION_OFF,
847 L1TF_MITIGATION_FLUSH_NOWARN,
848 L1TF_MITIGATION_FLUSH,
849 L1TF_MITIGATION_FLUSH_NOSMT,
850 L1TF_MITIGATION_FULL,
851 L1TF_MITIGATION_FULL_FORCE
852 };
853
854 extern enum l1tf_mitigations l1tf_mitigation;
855
856 enum mds_mitigations {
857 MDS_MITIGATION_OFF,
858 MDS_MITIGATION_FULL,
859 MDS_MITIGATION_VMWERV,
860 };
861
862 extern bool gds_ucode_mitigated(void);
863
864 #endif /* _ASM_X86_PROCESSOR_H */
865